Purpose: Transmission dynamic models have been used over the years for assessing the health and economic impact of interventions to control the spread of infectious diseases. Given the growing requirements to use probabilistic sensitivity analysis (PSA) to provide a single, global analysis of uncertainty in input parameters, use of PSA is hampered by the inherent complexity of these dynamic models. The objective of this study is to assess the feasibility of conducting PSA in a complex dynamic model by using HPV vaccination of boys and men in the United States as an example.
Method: We used previously developed age-structured mathematical population models of HPV vaccination. All 14 models consist of a series of nonlinear ordinary differential equations. Inputs for the models were obtained from public data sources, published literature, and analyses of clinical trial data. Inputs related to vaccine properties and uptake, cost, and quality of life weights were included in a PSA. Latin hypercube sampling techniques were used to generate 200 random samples. These were used as inputs in the simulations carried out in Mathematica. The uncertainty in the results was summarized in cost-effectiveness acceptability curves, and most influential parameters were determined using the partial rank correlation coefficient.
Result: We succeeded in performing the simulation of all models separately, calculating several measures of vaccine impact, and combining the results for cost-effectiveness analysis. For example, we found that compared with a program of vaccinating girls and women only, including boys and men has potentially substantial public health and economic benefits. The mean cost-effectiveness ratio of this strategy was $25,700 (Range: 13,600–48,800) per QALY gained, signifying cost effectiveness at the commonly cited thresholds. Influential parameters include vaccine cost, efficacy of only two doses, compliance, degree and duration of protection, quality of life of men with genital warts, and uptake among girls and women. However despite the ability to generate these findings, the amount of time needed to complete the 200 runs for a single model could be up to 17 hours depending on the complexity of individual models.
Conclusion: Although this study demonstrated the feasibility of conducting PSA in complex dynamic models, use of PSA may still be hampered with more complex models and/or a need to generate greater sets of runs (e.g., n=1,000).